VoiceXML

VoiceXML

Nuance Speech Analysis 92% of customer

service is through phone.

84% of industrialists believe speech better than web.

1st Qtr8%

2nd Qtr92%

1st Qtr16%

2nd Qtr84%

W3C (’02)

VoiceXML Forum (’00)

Motorola (’98)

HP (’98)

IBM (’98)

Bell/Lucent (’98)

AT&T (‘95)

History of VoiceXML

PML

PML

SpeechML

TalkML

VoxML

VoiceXML 1.0

VoiceXML 2.0

VoiceXML Open standard-language for serving

voice/audio documents.

VoiceXML is designed for creating audio dialogs that feature.

Synthesized speech, Digitized audio, Recognition of spoken and DTMF key input, Recording of spoken input, Telephony and Mixed-Initiative conversations.

VoiceXML (Cont’d) VoiceXML allows scripts/CGIs etc.

Can take input from the listener via speech(fill out forms like in HTML).

Used extensively for automated call handling.

Makes info accessible over (cell) phones

The next revolution on the Web.

Architectural Model

Goals of VoiceXML Web development and content delivery into

voice response applications.

Minimize client/server interactions.

Separate code from service logic.

Shield the application authors from platform specific details.

Voice Browser Software platform running on a network server.

It supports the following features. ASR DTMF Recognition grammars Mixed-initiative dialog TTS

Voice browser:VoiceXML :: Web browser:HTML

Voice Enabling

Sample VoiceXML Code <vxml version="2.0">

<form> <field name="rich" type="boolean">

<prompt>Would you like to get rich quick?</prompt> <filled>Gotcha.

<if cond="rich">You want to be rich! <goto next="rich.vxml" />

<else /> You don't want to be rich.

<goto next="poor.vxml" /> </if> </filled> </field> </form> </vxml>

Problem with VoiceXML Navigation of the voice document.

Author has to ask where listener will like to go next.

Listener has absolutely no control over navigation.

Tedium, Adv.Applications not possible.

Analogy: Scroll vs book

Our Architecture

Voice Anchors Speech labels that listeners can place on a

dialog.

Listener can return to that dialog later by uttering that label.

Hard to implement, as free-form speech recognition is not possible.

Need to incorporate in the voice browser.

Voice Anchors We developed a number of methods for

attaching voice anchors.

Most practical method: Spelling.

Anchor as a whole word.

Default anchors

Default navigation strategies

Cumulative Anchors Different dialogs can be marked with the same

label.

Recalling the label reads out the corresponding dialogs.

Multiple cumulative anchors in a single document.

Grammar Set of valid expressions.

Each dialog references one or more grammars.

Nuance Grammar Specification Language (GSL).

Inline grammar and Offline grammar. Offline provides the following advantages:

Can be generated dynamically (via Cgi’s, Asp's). Reused by multiple dialogs or applications. Updated and modified without change in source code.

Subgrammars and Form-level grammar.

Sample Grammar code<grammar type="application/x-gsl" mode="voice"><![CDATA[[[(skip)]{<option "skip">}[(previous)]{<option "previous">}[(place anchor) (call mark) (begin mark)]{<option

"mark">}[(recall mark) (recall anchor) (recall)]{<option "recall">}]]]>

Future work

Applications The Voice Web.

Talking books

Mathematics for visually impaired.

Hazardous Material Emergency Response.